README.Rmd

---
output:
  github_document:
    html_preview: false
editor_options: 
  chunk_output_type: console
bibliography: library.bib
---

```{r setup1, include=FALSE}
knitr::opts_chunk$set(fig.path = "man/figures/README_", warning = FALSE, message = FALSE, error = FALSE, echo = TRUE)

submission_to_cran <- TRUE
library(tidyverse)
```


# SCORPIUS

<!-- badges: start -->
[![R-CMD-check](/~https://github.com/rcannood/SCORPIUS/actions/workflows/R-CMD-check.yaml/badge.svg)](/~https://github.com/rcannood/SCORPIUS/actions/workflows/R-CMD-check.yaml)
[![CRAN_Status_Badge](https://www.r-pkg.org/badges/version/SCORPIUS)](https://cran.r-project.org/package=SCORPIUS)
[![Codecov test coverage](https://codecov.io/gh/rcannood/SCORPIUS/branch/master/graph/badge.svg)](https://app.codecov.io/gh/rcannood/SCORPIUS?branch=master)
<!-- badges: end -->

SCORPIUS an unsupervised approach for inferring linear developmental chronologies from single-cell
RNA sequencing data. In comparison to similar approaches, it has three main advantages:

* **It accurately reconstructs linear dynamic processes.** 
  The performance was evaluated using a quantitative evaluation pipeline and
  ten single-cell RNA sequencing datasets. 
  
* **It automatically identifies marker genes, speeding up knowledge discovery.**
  
* **It is fully unsupervised.** Prior knowledge of the relevant marker genes or 
  cellular states of individual cells is not required.
  
News: 

* See `news(package = "SCORPIUS")` for a full list of changes to the package.

* Our preprint is on [bioRxiv](https://biorxiv.org/content/early/2016/10/07/079509)
  [@Cannoodt2016].

* Check out our [review](https://dx.doi.org/10.1038/s41587-019-0071-9) on Trajectory Inference methods
  [@Saelens2019].
  
## Installing SCORPIUS

You can install:

* the latest released version from CRAN with

    ```R
    install.packages("SCORPIUS")
    ```

* the latest development version from GitHub with

    ```R
    devtools::install_github("rcannood/SCORPIUS", build_vignettes = TRUE)
    ```

If you encounter a bug, please file a minimal reproducible example on the [issues](/~https://github.com/rcannood/SCORPIUS/issues) page. 

## Learning SCORPIUS

To get started, read the introductory example below, or read one of the vignettes containing more elaborate examples:

```{r vignettes, results='asis', echo=FALSE}
walk(
  list.files("vignettes", pattern = "*.Rmd"),
  function(file) {
    title <- 
      read_lines(paste0("vignettes/", file)) %>% 
      keep(~grepl("^title: ", .)) %>% 
      gsub("title: \"(.*)\"", "\\1", .)
    vignette_name <- gsub("\\.Rmd", "", file)
    cat(
      "* ", title, ":  \n",
      "`vignette(\"", vignette_name, "\", package=\"SCORPIUS\")`\n", 
      sep = ""
    )
  }
)
```

## Introductory example

```{r, echo=F}
set.seed(1)
```

This section describes the main workflow of SCORPIUS without going in depth in the R code. For a more detailed explanation, see the vignettes listed below.

To start using SCORPIUS, simply write:
```{r libraries, message=FALSE}
library(SCORPIUS)
```


The `ginhoux` dataset [@Schlitzer2015] contains 248 dendritic cell progenitors in one of three cellular cellular states: MDP, CDP or PreDC. Note that this is a reduced version of the dataset, for packaging reasons. See ?ginhoux for more info.
```{r load_ginhoux}
data(ginhoux)
expression <- ginhoux$expression
group_name <- ginhoux$sample_info$group_name
```


With the following code, SCORPIUS reduces the dimensionality of the dataset and provides a visual overview of the dataset. 
In this plot, cells that are similar in terms of expression values will be placed closer together than cells with dissimilar expression values.

```{r reduce_dimensionality}
space <- reduce_dimensionality(expression, "spearman")
draw_trajectory_plot(space, group_name, contour = TRUE)
```


To infer and visualise a trajectory through these cells, run:
```{r infer_trajectory}
traj <- infer_trajectory(space)
draw_trajectory_plot(space, group_name, traj$path, contour = TRUE)
```

To identify candidate marker genes, run:
```{r find_tafs, message=F, warning=F}
# warning: setting num_permutations to 10 requires a long time (~30min) to run!
# set it to 0 and define a manual cutoff for the genes (e.g. top 200) for a much shorter execution time.
gimp <- gene_importances(
  expression, 
  traj$time, 
  num_permutations = 10, 
  num_threads = 8, 
  ntree = 10000,
  ntree_perm = 1000
) 
```

To select the most important genes and scale its expression, run:
```{r calculate_pvalue, message=F, warning=F}
gimp$qvalue <- p.adjust(gimp$pvalue, "BH", length(gimp$pvalue))
gene_sel <- gimp$gene[gimp$qvalue < .05]
expr_sel <- scale_quantile(expression[,gene_sel])
```


```{r echo=F}
# reverse the trajectory. This does not change the results in any way,
# other than the heatmap being ordered more logically.
# traj <- reverse_trajectory(traj)
```

To visualise the expression of the selected genes, use the `draw_trajectory_heatmap` function.
```{r visualise_tafs, fig.keep='first'}
draw_trajectory_heatmap(expr_sel, traj$time, group_name)
```

Finally, these genes can also be grouped into modules as follows: 
```{r moduled_tafs, fig.keep='first'}
modules <- extract_modules(scale_quantile(expr_sel), traj$time, verbose = F)
draw_trajectory_heatmap(expr_sel, traj$time, group_name, modules)
```

## References